wav2vec2-liepa-1-percent

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the COMMON_VOICE - LT dataset. It achieves the following results on the evaluation set:

• Loss: 0.5774
• Wer: 0.5079

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.23	100	3.3596	1.0
No log	0.46	200	2.9280	1.0
No log	0.69	300	1.5091	0.9650
No log	0.93	400	0.9943	0.9177
3.1184	1.16	500	0.7590	0.7793
3.1184	1.39	600	0.7336	0.7408
3.1184	1.62	700	0.7040	0.7618
3.1184	1.85	800	0.6815	0.7233
3.1184	2.08	900	0.6457	0.6865
0.7917	2.31	1000	0.5705	0.6813
0.7917	2.55	1100	0.5708	0.6620
0.7917	2.78	1200	0.5888	0.6462
0.7917	3.01	1300	0.6509	0.6970
0.7917	3.24	1400	0.5871	0.6462
0.5909	3.47	1500	0.6199	0.6813
0.5909	3.7	1600	0.6230	0.5919
0.5909	3.94	1700	0.5721	0.6427
0.5909	4.17	1800	0.5331	0.5867
0.5909	4.4	1900	0.5561	0.6007
0.4607	4.63	2000	0.5414	0.5849
0.4607	4.86	2100	0.5390	0.5587
0.4607	5.09	2200	0.5313	0.5569
0.4607	5.32	2300	0.5893	0.5797
0.4607	5.56	2400	0.5507	0.5954
0.3933	5.79	2500	0.5521	0.6025
0.3933	6.02	2600	0.5663	0.5989
0.3933	6.25	2700	0.5636	0.5832
0.3933	6.48	2800	0.5464	0.5919
0.3933	6.71	2900	0.5623	0.5832
0.3367	6.94	3000	0.5324	0.5692
0.3367	7.18	3100	0.5907	0.5394
0.3367	7.41	3200	0.5653	0.5814
0.3367	7.64	3300	0.5707	0.5814
0.3367	7.87	3400	0.5754	0.5429
0.2856	8.1	3500	0.5953	0.5569
0.2856	8.33	3600	0.6275	0.5394
0.2856	8.56	3700	0.6253	0.5569
0.2856	8.8	3800	0.5930	0.5429
0.2856	9.03	3900	0.6082	0.5219
0.2522	9.26	4000	0.6026	0.5447
0.2522	9.49	4100	0.6052	0.5271
0.2522	9.72	4200	0.5871	0.5219
0.2522	9.95	4300	0.5870	0.5236
0.2522	10.19	4400	0.5881	0.5131
0.2167	10.42	4500	0.6122	0.5289
0.2167	10.65	4600	0.6128	0.5166
0.2167	10.88	4700	0.6135	0.5377
0.2167	11.11	4800	0.6055	0.5184
0.2167	11.34	4900	0.6725	0.5569
0.1965	11.57	5000	0.6482	0.5429
0.1965	11.81	5100	0.6037	0.5096
0.1965	12.04	5200	0.5931	0.5131
0.1965	12.27	5300	0.5853	0.5114
0.1965	12.5	5400	0.5798	0.5219
0.172	12.73	5500	0.5775	0.5009
0.172	12.96	5600	0.5782	0.5044
0.172	13.19	5700	0.5804	0.5184
0.172	13.43	5800	0.5977	0.5219
0.172	13.66	5900	0.6069	0.5236
0.1622	13.89	6000	0.5850	0.5131
0.1622	14.12	6100	0.5758	0.5096
0.1622	14.35	6200	0.5752	0.5009
0.1622	14.58	6300	0.5727	0.5184
0.1622	14.81	6400	0.5795	0.5044

Framework versions

• Transformers 4.19.0.dev0
• Pytorch 1.10.0+cu113
• Datasets 2.1.0
• Tokenizers 0.12.1